Consistent iteration and multi-view transfer learning-based pedestrian re-identification method

Abstract

The present invention relates to a consistent iteration and multi-view transfer learning-based pedestrian re-identification method. The method comprises the following steps that: feature extraction is performed on pedestrian images, local and global multi-view image visual terms are obtained; a transfer learning method and a discriminatory analysis method are adopted to construct a consistent iteration and multi-view transfer learning optimization model, and the model is solved, so that middle-level image feature descriptors are obtained; calculation is performed based on obtained low-level feature descriptors and the middle-level feature descriptors, so that final multi-level image feature descriptors can be obtained; and a cross-view-based secondary discriminatory analysis method is utilized to measure the similarity of pedestrians, so that the similarity sequencing result of the pedestrian images can be obtained. Compared with the prior art, the consistent iteration and multi-view transfer learning-based pedestrian re-identification method of the invention has the advantages of high robustness and reliability under the change of factors such as illumination and rotation generated under a multi-view condition, can extract the bottom-level and middle-level feature descriptors of the images and has high pedestrian identification capability.

Description

technical field [0001] The invention relates to the field of intelligent monitoring video analysis, in particular to a pedestrian re-identification method based on consistent iterative multi-view transfer learning. Background technique [0002] The core problem of pedestrian re-identification (Re-ID) is to study how to accurately match the same pedestrian in a multi-camera non-overlapping system. Due to the influence of environmental factors such as viewing angle changes, lighting conditions, and posture changes, as well as human factors, the images of the same pedestrian captured by the camera are very different, which also makes the pedestrian re-identification problem challenging. At present, the solutions to this problem are mainly divided into two types: methods based on pedestrian appearance features and methods based on metric learning. Among them, the methods based on pedestrian appearance features mainly extract robust feature descriptors from pedestrian appearance...

AI Technical Summary

Problems solved by technology

Although these metric-based methods are better than existing person re-identification methods, they are still limited by some classic problems, such as: when the model is learning, the data distribution of multiple views is inconsistent, and the small sample size and so on

[0005] In traditional machine learning models, there is often an important premise, that is, assuming that there is consistency in the distribution of sample data between the training set and the test set. However, it is difficult to guarantee the establishment of this assumption in the data collected in real life.

Images